A vehicle trajectory prediction model that integrates spatial interaction and multiscale temporal features.

Area of Science:

• Artificial Intelligence
• Computer Vision
• Transportation Engineering

Background:

• Accurate real-time trajectory prediction is crucial for intelligent vehicles in heterogeneous traffic.
• Understanding spatial interactions and temporal dynamics is key to improving prediction accuracy.

Purpose of the Study:

• To develop a neural network model for precise real-time trajectory prediction of human-driven vehicles around intelligent vehicles.
• To integrate spatial interaction information with long-term and short-term time series characteristics.

Main Methods:

• A Graph Attention Network (GAT) encoder processes historical vehicle states and spatial interactions.
• A Transformer encoder extracts global dependencies, enhanced by residual connections.
• An LSTM encoder-decoder architecture captures short-term temporal features for trajectory generation.

Main Results:

• The proposed model demonstrated superior predictive performance compared to baseline models on a public dataset.
• Integration of GAT and Transformer architectures effectively captures complex vehicle interactions.
• LSTM component successfully models short-term temporal dependencies for accurate forecasting.

Conclusions:

• The developed neural network model offers a significant advancement in real-time trajectory prediction for heterogeneous traffic.
• The model's ability to integrate spatial and temporal features provides a robust solution for autonomous driving safety.
• Further research can explore real-world deployment and adaptation to diverse traffic scenarios.